Skip to content

Messages-style interface, native replies and reactions on any message: interested? - #156

Closed
marcus wants to merge 52 commits into
nixfred:mainfrom
marcus:messages-ui
Closed

marcus wants to merge 52 commits into
nixfred:mainfrom
marcus:messages-ui

Conversation

@marcus

@marcus marcus commented Nov 9, 2026

Copy link
Copy Markdown
Contributor

Hi Fred,

I've been running Blip daily and have built a Messages-style interface on top of it, along with a few features. Before I go any further I wanted to ask: would you be interested in any of this upstream? It's a big PR, so I'm opening it to start a conversation, not expecting a merge as-is. Happy to split it, drop parts, or keep the UI in my fork and send only the features you want.

It's merged with your main as of today, including #150–#153 and the Find My cities. bun test passes 899/899 and every CI step passes. All screenshots use the scripts/demo fixtures; no real conversations.

Before (your main) After

Interface

  • Layout: a contact grid and a centered photo-and-name header, with rounded glass controls and line icons. New Message sits in the conversation header, and the window is titled "Messages".
  • Reactions: emoji and sticker reactions use round badges with a thought-bubble tail.
  • Message times: a sideways swipe reveals them, replacing the hover tooltip (time-reveal.ts).
  • Notices: status and errors appear as small floating notices that dismiss themselves.
  • Backgrounds: optional per-conversation or default backgrounds, stored as metadata-stripped local copies.

Features (these could stand alone)

  • Native replies to text and single pictures, behind replies=on and using the new imsg-reply on the Mac. See docs/replies.md.
  • "Delivered" under your newest message until it's read.
  • Tapbacks and custom emoji reactions on any message on macOS 27, not just the newest. imsg-react uses the bubble's own Tapback Details action and takes about 10 seconds. See docs/reactions.md.
  • A searchable emoji picker in the composer and for reactions.
  • Stickers render without photo framing, attach to their target message and keep transparency.
  • Voice messages play inline, with waveforms and seeking.

Things it removes or changes that you may not want

  • Status/version banner: the app's banner is gone.
  • "mark all read" link: it's now Mark All as Read in the conversation menu, and the a key still works.
  • Header subtitle: it no longer shows the handle or "group · read-only (id unknown)". The shared city stays, as a caption under the name.
  • Default font: it now prefers Inter, then a Linux sans font, before the theme font. ui_font=theme brings back the old behaviour.
  • Mac tools: Mac users need to re-run bridge/mac/install.sh for the reply and reaction changes.
  • Version: manifest.json says 2.7.0, but the number is entirely your call.

If you'd rather not take any of it, no hard feelings. I'll keep the fork in sync with your main either way. Thanks for building Blip.

— Marcus

🤖 Generated with Claude Code

marcus added 30 commits November 6, 2026 14:09
Share the Unicode emoji catalogue between reaction and composer pickers. Preserve classic/custom identity and keep emoji payloads on bounded stdin. Extend the Mac bridge with exact Accessibility identity matching and type-plus-emoji receipt verification; refuse unavailable pickers without guessing. Preserve target, no-retry, consent, and message-content privacy invariants. Cover the protocol, AX safeguards, pending state, and real QML keyboard behavior using synthetic fixtures.
Recognize sticker metadata, retain standalone and orphan stickers, and fold associated media onto parent bubbles with bounded placement and activity timestamps. Preserve self-thread decoration and show explicit missing-media fallbacks.

Keep alpha through HEIC PNG conversion and bounded APNG/HEICS animated WebP previews. Validate raster signatures before rendering, preserve media-cache privacy and read-boundary invariants, and add synthetic SQLite, TypeScript, and real Qt animation checks.
Replace the redundant app banner with a centered contact photo/name, compact panes and rounded composer controls. Integrate emoji insertion and attributed sticker rendering with bounded previews and safe original-file actions. Add local PNG/JPEG backgrounds with hashed per-chat overrides in bridge.conf, atomic private writes, kernel save locking, and stale-result guards.

Preserve message-body stdin-only transport, group targeting, read marks and scroll selection. Validate with 840 Bun tests, Qt background callback tests, and synthetic wide/narrow/minimum-size screenshots; no real conversations are used.
Remove ImageMagick-generated orientation/time chunks from private background copies while retaining rendering chunks unchanged. Assert stored PNGs carry no EXIF or text metadata. Surface animation fallback to accessibility and pause every placed sticker layer outside the viewport.
Use public Accessibility APIs to retain picker context and discover detached emoji search results. Validate exact glyph identities, enabled state, stable bounds and global hit-test before one click; require exact chat.db receipts and never retry an uncertain action. Preserve draft/clipboard and focus invariants. Add synthetic origin, selection, and input guard tests; document verified self-thread add, replace, remove, and ZWJ behavior.
marcus and others added 14 commits November 6, 2026 17:48
CI's default font is taller, so a fixed height failed there. Compare a short
notice with a wrapped one instead. Only the test changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The nixfred GitHub account no longer resolves, so its name could be claimed by
someone else. Remove scripts/sync-bridge.sh, which downloaded Mac tools from it,
and the security-report link. Drop the upstream URL and product name from the
link preview User-Agent, which told every previewed site the client was Blip.

An audit found no telemetry, so no behavior other than the User-Agent changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It was pinned to the top of the header row while the conversation actions
button is vertically centered, so with a conversation selected the taller
header left the two at different heights. Centering it puts both in one row.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Own messages had no status until they were read, so the newest one sat
against the composer and the whole list jumped when "Read" appeared.

Your newest message now carries one fixed-height status line: Sending…,
then Delivered, then Read. The line is reserved at the end of the thread
before it has text. An older Read no longer stays above a newer Delivered,
because removing it later would shift the list too.

imsg reports delivery for own iMessage and RCS rows; carrier SMS has no
such state in Messages. A read from an earlier day shows only the day, and
Read is one weight heavier than its detail. The transcript also ends with
a small gap above the composer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A self-thread keeps the inbound echo and drops the sent row, which is the
only one that knows whether the message was delivered or read, so these
threads never showed a status. The kept row now takes both fields. The demo
bridge reports delivery too, so its pictures show the new status line.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The read-sync fixture builds its rows by hand and had no is_delivered, which
failed CI. It now has the column, and a new test covers which rows report
delivery.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Reply was only a quote, because AppleScript cannot reply. Messages' own
"Reply…" bubble action can be driven through Accessibility, so imsg-reply
presses it, fills the reply field directly and confirms in chat.db.

It covers plain text and single pictures in a 1:1, opt-in with replies=on,
and refuses what has not been measured. A composer draft stops the run,
because the reply view would take it. The text never touches the clipboard.
Exit 76 means Return was pressed but not confirmed, so Blip does not hand
the words back, which could send them twice.

Replies to a file now show it above the bubble, with a thumbnail for a
picture, where they used to show nothing. docs/replies.md records what was
measured, the dead ends and what the unbuilt cases would need.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A reply can land under another unmerged address of the same person, which
makes it look lost in Blip. The docs also warn about queries and captures
that caught the wrong thing during testing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Takes upstream's 15 commits since d51b0bc: Pin/Unpin on macOS 27,
green DMs no longer flipped back to iMessage, bar glyph sizing with
an Omarchy <4.5 fallback, imsg-read over ssh on macOS 27.2, and
opt-in Find My shared cities (nixfred#152).

Resolutions:
- CHANGE.log, CONTRIBUTORS-X.md, WEEKLY-NOTES.md, X-POSTS.md stay
  deleted; the fork removed them on purpose.
- BlipView.qml keeps the fork's avatar and name-pill header in place
  of PanelHero; the shared city shows as a caption under the name,
  and cityStatus() reads that caption.
- ci.yml and install.sh carry both imsg-reply and findmy_city.py.
- ui.test.ts keeps both sides' tests.
- DemoShell.qml: both sides added a takeTapbackFailure stub; one kept.

bun test 898/898; CI python steps pass; demo screenshots checked.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Upstream's list header had a "mark all read" link; the Messages-style
list dropped it and kept only the 'a' key. The action comes back as a
menu item under Mark as Read, behind the same canMarkAllRead gate
(push_read=off still keeps it local).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Reverts 68de47f. The nixfred account was unreachable on Nov 6 and is
back with the repository intact, so restore the security-advisory
link, the link-preview User-Agent, scripts/sync-bridge.sh and the
claude-on-mac pin.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The repository is going public and the original is back, so drop the
private-continuation note and docs/CONTINUATION.md, install from
nixfred/blip, fold the three 2.7.0-private changelog entries into one
2.7.0 section (adding native replies and Delivered), and update the
test badge to 899.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Synthetic data only (scripts/demo/blip-shots). The new shot shows
Mark All as Read in the conversation menu.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The fork removed these as not applying to it; a change offered
upstream should not delete the maintainer's own files.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@marcus
marcus requested a review from nixfred as a code owner November 9, 2026 17:16
@nixfred

nixfred commented Nov 10, 2026

Copy link
Copy Markdown
Owner

@marcus, thank you, Marcus. This is a lot of careful work, and asking first instead of dropping 17,000 lines was the right way to do it. The demo fixtures instead of real conversations, a green CI run and a branch already merged with today's main also made it easy to look at. I checked: gitleaks finds nothing, and it still merges cleanly on top of the three PRs merged today (b87c8a8).

Here is what I am deciding, and why.

The restyle stays in your fork. The contact grid, the glass controls, the new default font, the removed banner and "mark all read" link, and the per-conversation backgrounds are a different product look, and Blip aims at one opinionated iMessage look that follows the Omarchy theme, with its few settings as bridge.conf keys rather than an appearance system (the notes in CLAUDE.md, under "Configuration is bridge.conf keys", say why). I would rather you keep the look you like than have me half-take it.

The features I would like, as separate PRs. In this order, each on current main, each with its own tests and docs, so each can be read and verified on its own:

  1. "Delivered" under your newest message until it is read.
  2. Tapbacks and custom emoji reactions on any message on macOS 27.
  3. Native replies, with imsg-reply behind replies=on, off by default as you have it.
  4. The searchable emoji picker, stickers, and inline voice messages (these can be one PR each).

Why separate. Replies add a new Mac tool that sends messages and a change to the blip-dispatch allow-list. Reactions on any message drive Messages' own UI for about 10 seconds each. Both are the kind of change that gets read line by line and then run on a real macOS 27.2 gateway, and I can only do that for a PR small enough to hold in one head. I have not read all 17,000 lines of this one, and I will not pretend I did.

I am leaving this PR open until the first split PR arrives, then closing it with a link. The version number is mine to set, so leave manifest.json alone. Thank you for building on Blip.

nixfred added a commit that referenced this pull request Nov 10, 2026
…Pin on 27.2 hit the 9-pin cap

WHAT: CHANGE.log entries for #155, #157 and #154 (newest first) plus a PEOPLE
  line for Gerald LaMontagne (lama7, X none found, asked on #157); the roster
  row for lama7 and Ian Swope's merged count (22 -> 24); a WEEKLY-NOTES W46
  section (what people said, the three merges, #156 and Fred's decision on it,
  open handle questions).
WHY: the CHANGE.log, roster and weekly notes are what the Sunday X post is
  written from, and the rule is to add them the day the work merges.
LIVE RESULT RECORDED: on Fred's macOS 27.2 gateway the self-thread is 58th of
  496; the paging in #154 reached it and Pin fired, then Messages answered
  "You can pin up to 9 conversations" because the pin list was already full.
  Nothing was pinned, the pin list is unchanged, the alert was dismissed.
  So reaching and firing on 27.2 is shown; a pin landing there is not, and
  #132 stays open. imsg-read deployed to the gateway (sha256 6786da2e09fc ==
  main), plugin tapback files deployed to gus, shell restarted, status healthy.
FOLLOW-UP NOT DONE: imsg-read reports "clicked but the conversation is not
  pinned" and leaves the cap alert open; a message naming the 9-pin limit
  needs Fred's approval as a one-way change.

LR-T: docs, weekly-notes, change-log, pr-run, pin, live-test, macos-27
LR-D: blip
LR-K: PR-154, PR-155, PR-157, PR-156, issue-132, nine-pin-cap
LR-R: b87c8a8, 1bb8d45, eb9ae62

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BXN1v67b9f9VHGe3FhaQuD

marcus commented Nov 10, 2026 •

Copy link
Copy Markdown
Contributor Author

Thanks for spelling out the boundary and review order. I’ve split the requested features into six PRs, each independently based on current main (3795d64), with its own tests/docs and no manifest change:

  1. Delivered: Show Delivered under the newest sent message until read #158
  2. macOS 27 tapbacks/custom emoji: Support tapbacks on older messages and exact emoji on macOS 27 #159
  3. Opt-in native replies: Add opt-in native replies with confirmed-send safeguards #160
  4. Searchable composer emoji: Add a searchable composer emoji picker #161
  5. Stickers: Render native stickers and bounded animated previews #162
  6. Inline voice messages: Play voice messages inline with real waveforms and seeking #163

The restyle stays in my fork. The extractions keep your theme, controls, banner and mark-all-read link; sticker reactions use a compact attached strip rather than the fork’s reaction-balloon treatment. Screenshots and runtime checks use invented fixtures, generated art and tones. No new live messages or reactions were sent for the split; the docs distinguish prior self-thread verification from the synthetic extraction checks.

There are shared touch points (especially reactions/replies and stickers/audio). Each works on main by itself; I’ll handle the follow-on rebases as earlier PRs merge rather than leaving those conflicts for you. The emoji catalogue is licensed data shared identically by the reaction and composer pickers; most of their added lines are that data and the generated QML modules.

Validation update: all six PRs now have green GitHub CI, including the feature-specific Qt/Python checks and shellcheck. Gitleaks scanned all six new commits and found no leaks. Confirmed that none changes manifest.json or deletes an upstream file.

nixfred commented Nov 10, 2026

Copy link
Copy Markdown
Owner

Thanks Marcus, this is exactly what I asked for. Closing this one in favor of the six splits, #158 through #163. I'll work through them in that order, starting with #158. Replies and tapbacks will take longer because I want to test them on a real macOS 27.2 gateway first.


Generated by Claude Code

nixfred added a commit that referenced this pull request Nov 10, 2026
WHAT: Marcus Vorwaller's first standalone split from his #156. imsg --rich adds delivered (own iMessage or RCS row with is_delivered), decorate() gives the receipt to only the newest sent message (Delivered, then Read with the existing time label; failed, withdrawn and scheduled rows never claim it), and dedupeSelfEcho keeps delivered and read_at when a self-thread echo folds into its sent twin, in either order. Carrier SMS unchanged; an older bridge that omits delivered keeps showing only Read.

WHY: iMessage shows Delivered under your newest message and Blip showed nothing until it was read. The old rule put Read under the newest READ message, which a later unread send left stale; the receipt now belongs to the newest sent message only (test changed to say so).

CONTEXT: door classifier one-way (bridge/mac/imsg); Fred approved from the literal diff on 10 Nov. Independent review found nothing. One conflict, test_read_sync.py (both new test classes kept). Needs the gateway's imsg updated for Delivered to appear.

LR-T: delivery-receipt, contributor-pr, marcus
LR-D: blip, thread, bridge-mac
LR-K: delivered-receipt, sentReceipt, newest-sent-owns-receipt
LR-R: e9adb10, 401b99d

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BXN1v67b9f9VHGe3FhaQuD
nixfred added a commit that referenced this pull request Nov 10, 2026
WHAT: Marcus Vorwaller's fourth standalone split from his #156. A smiley button beside Send opens a searchable Unicode catalogue (names, nine categories, an exact pasted sequence always first, keyboard navigation). Choosing an emoji replaces the selection or inserts at the caret and returns focus to the composer without sending; Escape keeps the draft. The catalogue is a generated 327 KB module (EmojiPicker.mjs) from licensed data (Unicode and Beeper/Automattic MIT notices are in docs/). The grid is a GridView over a JS array, so only visible cells are built.

WHY: the composer had no way to browse or search for an emoji; one had to be pasted in from somewhere else.

CONTEXT: door classifier one-way only for a 4-line CI step (no bridge, send path or permission); Fred approved on 10 Nov. Rendered in a clean Test Drive VM on invented data: themed popup, search, category list, grid. One cell showed an empty box for an emoji newer than that VM's font (a font gap, noted for Marcus). Qt UI test passes offscreen (6).

LR-T: emoji-picker, contributor-pr, marcus, composer
LR-D: blip, composer, ui
LR-K: composer-emoji-picker, emoji-catalogue, popup
LR-R: 2a2060d, 401b99d

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BXN1v67b9f9VHGe3FhaQuD
nixfred added a commit that referenced this pull request Nov 10, 2026
…views

WHAT: Marcus Vorwaller's fifth standalone split from his #156. iMessage stickers render as stickers: placed on the message they were dragged onto (position, scale and rotation from Messages' own metadata), standalone when they stand alone, with Memoji, Genmoji and app attribution in the accessible name. Animated GIF, APNG and WebP play inline (an APNG is converted to a lossless animated WebP, because Qt draws APNG as a still); Live Stickers (HEICS) become animated WebP through ffmpeg, or the Mac's transparent still when ffmpeg is missing. Bridge: imsg --rich carries a validated `sticker` object and a per-message `stickers` overlay list, imsg chats flags a sticker as a chat's last attachment, imsg attachment --png keeps transparency. blip-setup names the optional Qt WebP plugin.
WHY: a sticker was an empty bubble or a made-up star tapback; people send them constantly.
CONTEXT: door classifier one-way (bridge/mac/imsg, scripts/blip-setup, .github/workflows/ci.yml); Fred approved the literal diff on 10 Nov. My independent review sent six points back; Marcus addressed each and rebased, and I re-verified rather than trusting the reply: the ffconcat-as-video exploit I reproduced on the first head is now refused on all three decode paths (ffprobe and ffmpeg are pinned to -f mov and the APNG decoder to -f apng before -i), Stickers.mjs is byte-identical to a fresh build of stickers.ts, 862 bun, 21 Python and 6 Qt tests pass, CI is green on 4c0ba92, and against the real gateway chat.db the new enrich SQL ran clean on 90 windows and the chat list is identical apart from the new sticker key. Not verified: a real sticker rendered from the Mac (none is the newest attachment in any chat). Residual risk, asked of Marcus as a follow-up: ffmpeg now decodes sender-supplied video stickers on this machine without a click (bounded to 8 streams, 2048 px, a 64M-pixel budget and 30 s; external track references are off by default); run it sandboxed.
TAGS: stickers, memoji, genmoji, apng, heics, ffmpeg, contributor-pr, marcus

LR-T: stickers, contributor-pr, marcus, media
LR-D: blip, bridge-mac, fetch, ui
LR-K: native-stickers, sticker-overlay, ffmpeg-bounded-decode, apng-to-webp
LR-R: ea64c98, 0d302b1, 401b99d

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BXN1v67b9f9VHGe3FhaQuD
@nixfred

nixfred commented Nov 11, 2026

Copy link
Copy Markdown
Owner

Closing this as I said on the 10th and forgot to do: it is replaced by the six splits. Four are merged (#158, #161, #162, #163), and #159 and #160 are in review. The restyle stays in your fork, as agreed. Thanks again for splitting it up so cleanly.


Generated by Claude Code

@nixfred nixfred closed this Nov 11, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants